Predicting an unobserved driver of regime shifts in social-ecological systems with universal dynamic equations

Kunal J. Rathore , John H. Buckner, Zechariah D. Meunier, Jorge Arroyo Esquivel, James R. Watson

Ecosystems around the world are anticipated to undergo regime shifts as temperatures rise and other climatic and anthropogenic perturbations erode the resilience of present-day states. Forecasting these nonlinear ecosystem dynamics can help stakeholders better prepare. One major challenge though is that regime shifts can be difficult to predict when they are driven by unobserved factors. In this paper, we advance Scientific Machine Learning methods, specifically Universal Dynamic Equations (UDEs), to identify changes in an unobserved bifurcation parameter that is driving an ecosystem regime shift. We demonstrate this approach using simulated data created from a dynamic model of a species population experiencing loss due to extraction or harvest that is unobserved. This could be, for example, illegal fishing from a fishery or unreported poaching in a game reserve. We show that UDEs can accurately identify changes in the slowly increasing harvest rate (the bifurcation parameter), and predict when a regime shift will occur. Compared to alternative forecasting methods, our UDE approach provides relatively accurate short-term predictions, and provides a new set of methods for ecosystem stakeholders and to manage fast-paced ecosystem change likely to happen in the coming decades.